Table 4 The table shows the test MAE of the SC model, proposed TL model and % error change for each of the target materials properties for prediction task of ‘JARVIS-2D Database’.

From: Structure-aware graph neural network based deep transfer learning framework for enhanced predictive analytics on diverse materials datasets

Property

Data Size

Base

MAE of SC Model

MAE of Proposed TL Model

% Error Change

BgOptb (eV)

1074

0.9991

0.3537

0.3449

-2.49

Deltae (eVatom−1)

1074

0.5972

0.1264

0.0715

-43.43

KLU (Å)

1073

17.673

9.6951

9.5837

-1.15

Magout (μB)

1072

1.1681

0.1951

0.2100

7.64

Encut (eV)

1070

153.22

71.511

81.200

13.55

Magoszi (μB)

1036

1.0969

0.2138

0.1564

-26.85

Epsx

885

7.7160

4.0413

2.7412

-32.17

Epsy

885

8.2292

3.8451

3.6114

-6.08

Epsz

885

2.2659

1.9724

0.9043

-54.15

PPF (μWm−1K−2)

800

406.56

297.18

288.27

-3.00

NPF (μWm−1K−2)

800

354.84

306.03

270.52

-11.60

Exfoli (meVatom−1)

742

59.268

37.131

36.213

-2.47

NSB (μVK−1)

733

82.250

54.303

48.856

-10.03

PSB (μVK−1)

707

101.23

51.462

51.007

-0.88

Spillage

602

0.3039

0.1899

0.1971

3.79

Pem300k (m0)

264

0.6613

0.7691

0.1417

-81.58

Nem300k (m0)

253

0.6801

0.5672

0.3964

-30.11

Mepsz

246

3.7593

1.3816

1.8648

34.97

Mepsx

244

11.003

10.3305

9.4178

-8.84

SLME (%)

244

11.335

6.3130

5.1177

-18.93

Mepsy

243

12.302

6.7274

6.4251

-4.49

ETC11 (GPa)

224

39.869

37.327

35.990

-3.58

ETC22 (GPa)

223

52.439

45.719

42.894

-6.18

BgMbj (eV)

217

1.6573

0.9177

0.7503

-18.24

ETC12 (GPa)

212

17.352

12.654

14.573

15.17

ETC44 (GPa)

206

6.1253

2.4420

2.4040

-1.56

MaxM (cm−1)

184

223.29

78.586

66.534

-15.34

MinM (cm−1)

183

19.081

49.546

1.937

-96.09

ETC55 (GPa)

180

2.0985

2.9595

0.1973

-93.33

ETC66 (GPa)

180

1.7033

1.4323

0.1185

-91.73

ETC13 (GPa)

153

1.8649

1.4365

0.9880

-31.22

ETC33 (GPa)

135

6.2401

7.9269

6.0764

-23.34

  1. The lowest MAE values in each row are highlighted in bold.